Pipeline internal detection method and system based on multi-sensor fusion

Abstract

The present invention relates to the technical field of pipeline data processing and health status assessment, and discloses a pipeline internal detection method and system based on multi-sensor fusion. The method includes: calculating the health index of pipeline status assessment data to obtain pipeline health status indicators, performing risk prediction on pipeline health status indicators to obtain risk levels; performing pipeline defect feature identification to obtain defect feature data, performing multi-category classification processing on defect feature data to obtain defect type data; performing precise position analysis on defect type data to obtain defect spatial distribution data and perform structural integrity assessment to obtain defect severity quantification data set; performing multi-criteria decision analysis on defect severity quantification data set and risk level to obtain defect repair priority ranking data, performing intelligent repair plan generation on defect repair priority ranking data to obtain repair suggestion data. The present invention can achieve a more comprehensive and accurate assessment of pipeline status.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline data processing and health status assessment, and in particular to a pipeline internal detection method and system based on multi-sensor fusion. Background Art

[0002] With the acceleration of urbanization and the aging of infrastructure, the safe operation of urban drainage pipeline systems faces severe challenges. Traditional pipeline inspection methods mainly rely on manual visual inspection and single sensor technology, such as closed-circuit television (CCTV) inspection, acoustic inspection, etc. These methods can detect obvious defects inside the pipeline to a certain extent, such as cracks, corrosion, and sediments. However, due to the complex environment and poor inspection conditions of urban drainage pipelines, a single inspection method is often difficult to fully and accurately assess the pipeline condition.

[0003] The main deficiencies of existing technologies include: insufficient accuracy and completeness of detection data, making it difficult to capture minor defects and potential risks inside the pipeline; lack of effective fusion and analysis capabilities for multi-source heterogeneous data, resulting in an incomplete assessment of the overall condition of the pipeline; inability to accurately quantify the health status of the pipeline and dynamically predict the risk level; low intelligence in defect identification and classification, with much manual intervention; lack of accurate assessment of pipeline structural integrity and scientific decision-making support for repair priorities. These problems have seriously restricted the effective management and maintenance of urban drainage pipeline systems. Summary of the invention

[0004] In order to solve the above-mentioned deficiencies in the prior art, an embodiment of the present invention provides a pipeline internal detection method and system based on multi-sensor fusion.

[0005] The present invention provides a pipeline internal detection method based on multi-sensor fusion, comprising: preprocessing raw data collected by multiple sensors to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set; extracting and fusing features on the registered data set to obtain a target feature vector, and performing multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data; calculating a health index on the pipeline status assessment data to obtain a pipeline health status indicator, and performing risk prediction on the pipeline health status indicator to obtain a risk level; and performing pipeline defect prediction on the pipeline health status indicator based on the risk level. The defect feature recognition is performed to obtain defect feature data, and the defect feature data is subjected to multi-category classification processing to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data; the defect type data is accurately analyzed for position through a spatial positioning algorithm to obtain defect spatial distribution data, and a structural integrity assessment is performed on the defect spatial distribution data to obtain a defect severity quantification data set; a multi-criteria decision analysis is performed on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and an intelligent repair plan is generated for the defect repair priority ranking data to obtain repair suggestion data.

[0006] The present invention also provides a pipeline internal detection system based on multi-sensor fusion, comprising:

[0007] A processing module, used for preprocessing the raw data inside the pipeline collected by multiple sensors to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set;

[0008] A fusion module is used to extract and fuse the features of the registration data set to obtain a target feature vector, and perform multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data;

[0009] A prediction module is used to calculate the health index of the pipeline status assessment data to obtain the pipeline health status index, and to perform risk prediction on the pipeline health status index to obtain the risk level;

[0010] A classification module, used to identify pipeline defect characteristics for the pipeline health status indicator based on the risk level to obtain defect characteristic data, and to perform multi-category classification processing on the defect characteristic data to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data;

[0011] An analysis module is used to perform precise position analysis on the defect type data through a spatial positioning algorithm to obtain defect spatial distribution data, and to perform structural integrity assessment on the defect spatial distribution data to obtain a defect severity quantification data set;

[0012] A generation module is used to perform a multi-criteria decision analysis on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and to generate an intelligent repair plan for the defect repair priority ranking data to obtain repair suggestion data.

[0013] In the technical solution provided by the present invention, the raw data collected by multiple sensors inside the pipeline are preprocessed and time-space registered, ensuring the consistency of data from different sensors in time and space, laying a solid foundation for subsequent analysis. This step effectively improves the data quality and reduces the errors caused by sensor differences. Secondly, by extracting and fusion of the registered data set, the target feature vector is obtained, and multi-level fusion analysis is performed, which greatly enhances the comprehensive evaluation ability of the pipeline status. This multi-dimensional feature analysis makes the pipeline status assessment more comprehensive and accurate, and can capture subtle changes that may be ignored by a single sensor. Then, the health index calculation and risk prediction of the pipeline status assessment data not only provide quantitative indicators of the current pipeline health status, but also predict possible risks in the future, providing an important basis for preventive maintenance. This step significantly improves the foresight and initiative of pipeline management. In addition, based on the risk level, the defect feature recognition and multi-category classification processing of the pipeline health status indicators are carried out, realizing the accurate identification and classification of different types of defects, including cracks, corrosion, deposition and structural deformation. This detailed classification helps to formulate targeted repair strategies and improve maintenance efficiency. The spatial positioning algorithm is used to accurately analyze the defect type data and conduct a structural integrity assessment, which not only accurately locates the defect location, but also assesses the impact of the defect on the overall structure of the pipeline, providing key information for maintenance decisions. Finally, a multi-criteria decision analysis is performed on the defect severity quantitative data set and risk level, and an intelligent repair plan is generated, realizing intelligent management of the entire process from detection to repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1It is a flow chart of a pipeline internal detection method based on multi-sensor fusion in an embodiment of the present invention;

[0016] Figure 2 Schematic diagram of a pipeline internal detection system based on multi-sensor fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0019] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0020] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 is a flow chart of a pipeline internal detection method based on multi-sensor fusion according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0021] S101, preprocessing the raw data inside the pipeline collected by multiple sensors to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set;

[0022] S102, extracting and fusing features of the registration data set to obtain a target feature vector, and performing multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data;

[0023] S103, calculating the health index of the pipeline status assessment data to obtain the pipeline health status index, and performing risk prediction on the pipeline health status index to obtain the risk level;

[0024] S104, identifying pipeline defect characteristics for pipeline health status indicators based on risk levels to obtain defect characteristic data, and performing multi-category classification processing on the defect characteristic data to obtain defect type data including crack data, corrosion data, deposition data, and structural deformation data;

[0025] S105, performing precise position analysis on the defect type data through a spatial positioning algorithm to obtain defect spatial distribution data, and performing structural integrity assessment on the defect spatial distribution data to obtain a defect severity quantification data set;

[0026] S106. Perform multi-criteria decision analysis on the defect severity quantification data set and risk level to obtain defect repair priority ranking data, and generate intelligent repair plans for the defect repair priority ranking data to obtain repair suggestion data.

[0027] It should be noted that the raw data collected by multiple sensors in the pipeline are preprocessed to obtain a multi-source data set through steps such as data cleaning, noise filtering, standardization, and segmentation. Different types of sensor data will undergo specific processing. For example, for visual sensor data, image enhancement and deblurring may be performed; for acoustic sensor data, spectrum analysis and filtering may be required. Subsequently, the segmented data set is transformed into the frequency domain using the Fourier transform algorithm, the characteristic frequency bands are extracted, and the timestamps are aligned and the spatial coordinates are transformed. This step ensures the consistency of data from different sensors in time and space. Finally, the missing data is supplemented by the interpolation algorithm, and outlier detection is performed to obtain a high-quality registered data set. Feature extraction and fusion are performed on the registered data set. This process includes multi-scale decomposition, principal component analysis, nonlinear mapping, and dimensionality reduction. Multi-scale decomposition can capture the characteristics of data at different scales, while principal component analysis helps to reduce the data dimension and retain the main information. Deep feature extraction is performed through the convolutional neural network algorithm, which can automatically learn the high-level feature representation of the data. Feature selection and weight calculation are then performed to ensure that the final feature vector contains the most relevant and important information. The fuzzy inference algorithm is applied to the fused feature vector for multi-level fusion analysis. This method can handle the uncertainty and ambiguity in the data. After confidence evaluation and threshold screening, reliable pipeline status assessment data is finally obtained.

[0028] The health index of pipeline status assessment data is calculated by first normalizing parameters and assigning weights, which ensures the comparability between different indicators. Then a health index system is constructed and the indicators are quantified. This system may include the structural integrity, material properties, operating conditions and other aspects of the pipeline. A clear health index is obtained through fuzzy comprehensive evaluation and clarification processing, which can intuitively reflect the overall health status of the pipeline. The gray prediction algorithm is used for time series analysis, which is particularly suitable for processing incomplete information systems. By setting risk thresholds, comparative analysis, grade division and knowledge correction, a scientific and reasonable risk level assessment result is finally obtained. Based on the risk level, the pipeline health status indicators are used to identify pipeline defect features. This step first performs correlation analysis and abnormal pattern recognition to find potential problem areas. Then the wavelet transform algorithm is used for time-frequency analysis and peak detection. This method can effectively identify abnormal features in the signal. The peak feature data is subjected to morphological processing and texture analysis to further extract the detailed features of the defects. The support vector machine algorithm is used for multidimensional feature extraction and cluster analysis. This method performs well in processing high-dimensional data and nonlinear classification problems. Finally, after feature matching, pattern recognition and multi-category labeling, detailed defect type data including crack data, corrosion data, deposition data and structural deformation data are obtained.

[0029] The spatial positioning algorithm is used to accurately analyze the position of defect type data. The specific steps include using the triangulation algorithm to calculate spatial coordinates. This method can determine the precise three-dimensional position of the defect through the data of multiple sensors. Pipeline geometric mapping is performed to map the defect position to the actual structure of the pipeline. Through density analysis and hot spot analysis, the concentrated area of ​​the defect is identified. The minimum enclosing rectangle algorithm is used to extract the boundary and calculate the area to accurately quantify the scope of the defect. Depth measurement and volume estimation are performed to comprehensively evaluate the severity of the defect. Finally, the finite element analysis algorithm is used to perform stress simulation and strength evaluation. This method can predict the impact of defects on the strength of the pipeline structure. Through safety factor calculation and quantitative classification, a quantitative data set of defect severity is finally obtained. Finally, a multi-criteria decision analysis is performed on the quantitative data set of defect severity and risk level. Including data fusion, weight allocation, criterion system construction and consistency verification to ensure the scientificity and rationality of the decision-making process. Clear priority data is obtained through fuzzy comprehensive evaluation and clarification processing, and priority sorting and threshold segmentation are performed using a sorting algorithm. Repair technology matching and resource demand analysis are performed on the graded priority data, considering existing technical conditions and resource constraints. Ultimately, through cost-benefit analysis and solution optimization, the best repair recommendation data is obtained, providing a scientific basis for pipeline maintenance decisions.

[0030] By executing the above steps, the raw data collected by multiple sensors inside the pipeline are preprocessed and time-space registered, ensuring the consistency of data from different sensors in time and space, laying a solid foundation for subsequent analysis. This step effectively improves data quality and reduces errors caused by sensor differences. Secondly, by extracting and fusion of the registered data set, the target feature vector is obtained, and multi-level fusion analysis is performed, which greatly enhances the comprehensive assessment capability of the pipeline status. This multi-dimensional feature analysis makes the pipeline status assessment more comprehensive and accurate, and can capture subtle changes that may be ignored by a single sensor. Then, the health index calculation and risk prediction of the pipeline status assessment data not only provide quantitative indicators of the current pipeline health status, but also predict possible risks in the future, providing an important basis for preventive maintenance. This step significantly improves the foresight and initiative of pipeline management. In addition, defect feature recognition and multi-category classification processing of pipeline health status indicators based on risk levels are performed to achieve accurate identification and classification of different types of defects, including cracks, corrosion, deposition, and structural deformation. This detailed classification helps to formulate targeted repair strategies and improve maintenance efficiency. The spatial positioning algorithm is used to accurately analyze the defect type data and conduct a structural integrity assessment, which not only accurately locates the defect location, but also assesses the impact of the defect on the overall structure of the pipeline, providing key information for maintenance decisions. Finally, a multi-criteria decision analysis is performed on the defect severity quantitative data set and risk level, and an intelligent repair plan is generated, realizing intelligent management of the entire process from detection to repair.

[0031] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0032] (1) Clean the raw data collected by multiple sensors inside the pipeline to obtain cleaned data, and filter the cleaned data to obtain filtered data;

[0033] (2) Performing data standardization on the filtered data to obtain standardized data, and performing data segmentation on the standardized data to obtain a segmented data set;

[0034] (3) The segmented data set is converted into frequency domain using the Fourier transform algorithm to obtain frequency domain data, and characteristic frequency bands are extracted from the frequency domain data to obtain characteristic frequency band data;

[0035] (4) Performing time stamp alignment on the characteristic frequency band data to obtain aligned data, and performing spatial coordinate transformation on the aligned data to obtain unified coordinate system data;

[0036] (5) Supplement the unified coordinate system data through the interpolation algorithm to obtain a complete data set, and perform quality assessment on the complete data set to obtain a candidate multi-source data set;

[0037] (6) Perform outlier detection on the candidate multi-source dataset to obtain a detected dataset, and perform spatiotemporal consistency verification on the detected dataset to obtain a target registration dataset.

[0038] Specifically, first, the raw data collected by multiple sensors inside the pipeline is cleaned. This includes identifying and processing missing values, outliers, and inconsistent data. For example, for ultrasonic sensor data, it may be necessary to remove zero values ​​or abnormally high values ​​caused by sensor failure. The cleaned data more accurately reflects the actual condition of the pipeline. The cleaned data is noise filtered to obtain filtered data. Noise filtering is an important step to improve data quality, and digital filtering technology is usually used. Different filtering methods may be required for different types of sensor data. For example, for vibration sensor data, a low-pass filter can be used to remove high-frequency noise; for image data, a median filter can be used to remove salt and pepper noise. The signal-to-noise ratio of the filtered data is significantly improved, laying the foundation for subsequent analysis. The filtered data is standardized to obtain standardized data. Standardization is the process of converting data of different dimensions and ranges to the same scale, usually using z-score standardization or minimum-maximum standardization methods. This step ensures that data from different sensors can be fairly compared and fused. The standardized data is then segmented to obtain a segmented data set. Data segmentation is usually based on time windows or spatial regions, with the goal of dividing a continuous data stream into manageable and analyzable data chunks.

[0039] The segmented data set is transformed into the frequency domain by the Fourier transform algorithm to obtain frequency domain data. Fourier transform is a mathematical tool that converts time domain signals into frequency domain representation. It is particularly suitable for analyzing periodic signals and discovering hidden frequency features. For pipeline detection, frequency domain analysis can reveal important information such as pipeline vibration and acoustic characteristics. Feature frequency bands are extracted from the frequency domain data to obtain feature frequency band data. Feature frequency bands usually contain the most relevant and diagnostically valuable information. For example, certain specific frequencies may be related to pipeline cracks or corrosion. The feature frequency band data is then timestamped to obtain aligned data. Timestamp alignment is a key step in multi-sensor fusion, ensuring that data from different sensors are synchronized in time. This is crucial for accurately analyzing pipeline status, especially when analyzing dynamic changes or event sequences. The aligned data is then transformed into a spatial coordinate system to obtain unified coordinate system data. Spatial coordinate transformation maps data collected by different sensors into the same spatial reference system, which is very important for accurately locating pipeline defects and performing spatial correlation analysis.

[0040] The unified coordinate system data is supplemented by interpolation algorithms to obtain a complete data set. Interpolation is an effective way to deal with missing data or uneven sampling. Common interpolation algorithms include linear interpolation and spline interpolation. For pipeline detection, interpolation can fill in areas where sensors are not well covered and provide more comprehensive pipeline status information. The complete data set undergoes quality assessment to obtain a candidate multi-source data set. Quality assessment includes consistency checks, integrity verification, etc. to ensure the reliability and representativeness of the data set. Finally, outlier detection is performed on the candidate multi-source data set to obtain a post-detection data set. Outlier detection aims to identify data points that are significantly different from the overall data distribution. These outliers may represent potential problems with the pipeline or sensor failures. The detected data set is then subjected to a spatiotemporal consistency check to obtain the final target registration data set. Spatiotemporal consistency checks ensure the coherence of data in time and space, and are an important guarantee of data credibility.

[0041] For example, suppose that ultrasonic sensors, vibration sensors, and image sensors are deployed in a 100-meter-long urban drainage pipe. The ultrasonic sensor collects data every 10 centimeters to measure the thickness of the pipe wall; the vibration sensor records the vibration spectrum every 1 meter; and the image sensor takes an image of the inner wall of the pipe every 5 meters. In the original data, there are 5% abnormal values ​​in the ultrasonic data (such as 0mm thickness), and these data are cleaned by the median replacement method. There is high-frequency noise in the vibration data, which is filtered using a low-pass filter with a cutoff frequency of 500Hz. The image data is denoised and enhanced. All data are standardized so that the data range of different sensors is unified to the [-1, 1] interval. The data segmentation uses 10 meters as a unit to obtain 10 data segments. Fast Fourier transform is performed on each data segment to extract the characteristic frequency band of 0-100Hz. Timestamp alignment ensures that the acquisition time error of the three sensor data in each 10-meter segment does not exceed 1 second. The spatial coordinate transformation maps all data to a one-dimensional coordinate system with the starting point of the pipe as the origin. For the vibration data, linear interpolation was used to estimate the vibration value at every 10 cm point to make it consistent with the spatial resolution of the ultrasonic data. Quality assessment removed outliers in 2 data segments. The resulting target registration dataset contains 8 high-quality 10-meter data segments, each containing aligned ultrasonic, vibration, and image data.

[0042] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0043] (1) Perform multi-scale decomposition on the registered data set to obtain multi-scale feature data, and perform principal component analysis on the multi-scale feature data to obtain the main feature data;

[0044] (2) Perform nonlinear mapping on the main feature data to obtain high-dimensional feature data, and perform dimensionality reduction processing on the high-dimensional feature data to obtain feature data after dimensionality reduction;

[0045] (3) Perform deep feature extraction on the feature data after dimensionality reduction through a convolutional neural network algorithm to obtain deep feature data, and perform feature selection on the deep feature data to obtain key feature data;

[0046] (4) Calculate the feature weight of the key feature data to obtain feature weight data, and perform weighted fusion on the feature weight data and the key feature data to obtain a fused feature vector;

[0047] (5) Perform multi-level analysis on the fused feature vector through fuzzy inference algorithm to obtain multi-level analysis results, and perform confidence evaluation on the multi-level analysis results to obtain confidence data;

[0048] (6) Perform threshold screening on the confidence data to obtain the screened evaluation data, and perform comprehensive scoring on the screened evaluation data to obtain the pipeline status evaluation data.

[0049] Specifically, multiscale decomposition is a technique that decomposes signals or data into different scales or resolutions. Common methods include wavelet transform and pyramid decomposition. This decomposition can capture the characteristics of data at different scales, which helps to identify subtle changes in pipeline status and large-scale structures. Multiscale feature data contains information about the original data at different frequencies or spatial scales, providing rich feature representations for subsequent analysis. Principal component analysis (PCA) is performed on the multiscale feature data to obtain the main feature data. PCA is a dimensionality reduction technique that projects the original features onto a set of orthogonal principal components through linear transformation. In pipeline detection, PCA can effectively reduce data redundancy and extract the most representative features. The main feature data retains most of the variability of the original data, while significantly reducing the data dimension and improving the efficiency of subsequent processing.

[0050] The main feature data is nonlinearly mapped to obtain high-dimensional feature data. Nonlinear mapping aims to capture complex nonlinear relationships in the data. Commonly used methods include kernel function mapping and neural network autoencoders. This step can transform the originally linearly inseparable features into a new feature space, making complex pipeline state patterns easier to identify. Although the dimension of high-dimensional feature data increases, it contains richer nonlinear feature information. Subsequently, the high-dimensional feature data is subjected to dimensionality reduction processing to obtain the reduced-dimensional feature data. This step usually uses nonlinear dimensionality reduction techniques, such as t-SNE or UMAP, to retain local structures and global relationships in high-dimensional space. The reduced-dimensional feature data greatly reduces the data dimension while retaining key information, which is convenient for subsequent deep learning processing.

[0051] The deep feature data after dimensionality reduction is extracted through the convolutional neural network (CNN) algorithm to obtain deep feature data. CNN is particularly suitable for processing data with spatial or temporal correlation, such as images or time series data. In pipeline detection, CNN can automatically learn complex feature representations, such as crack textures, corrosion patterns, etc. The deep feature data contains high-level abstract features, which are essential for identifying complex pipeline conditions. Feature selection is performed on the deep feature data to obtain key feature data. The feature selection process uses methods such as L1 regularization or random forest feature importance to identify the most relevant and discriminative features. The key feature data contains the most important feature subset for pipeline status assessment, which not only improves the interpretability of the model but also reduces the risk of overfitting.

[0052] The feature weights of the key feature data are calculated to obtain feature weight data. The weight calculation takes into account the contribution of each feature to the pipeline status assessment, and methods such as information gain or Fisher score can be used. The feature weight data reflects the relative importance of different features in the assessment process. Subsequently, the feature weight data and the key feature data are weighted and fused to obtain a fused feature vector. This step integrates feature information from different sources into a unified representation through weighted summation or more complex fusion algorithms. The fused feature vector is analyzed at multiple levels using a fuzzy reasoning algorithm to obtain a multi-level analysis result. Fuzzy reasoning can handle uncertainty and ambiguity in data, and is particularly suitable for handling complex decision-making problems in pipeline status assessment. The multi-level analysis results include fuzzy assessments of pipeline status, such as "slight corrosion" and "moderate cracks". The confidence assessment of the multi-level analysis results is performed to obtain confidence data. The confidence assessment measures the reliability of each analysis result and can be calculated based on the membership function value of fuzzy reasoning or statistical methods. The confidence data provides important reference information for subsequent decision-making.

[0053] Finally, the confidence data is threshold screened to obtain the screened evaluation data. The threshold screening eliminates the low-confidence evaluation results and retains the high-confidence pipeline status evaluation. The screened evaluation data is comprehensively scored to obtain the final pipeline status evaluation data. The comprehensive score takes into account multiple factors, such as defect severity, confidence, and potential risks, and provides a quantitative assessment of the overall status of the pipeline.

[0054] For example, suppose a 100-meter-long urban drainage pipe is inspected using three sensors: ultrasonic, vibration, and image. First, the registered dataset is decomposed by 4-level wavelet to obtain multi-scale feature data, including 16 feature subbands from high frequency to low frequency. Through PCA analysis, the first 10 principal components are selected, which explain 95% of the data variance. The 10 principal components are kernel mapped to expand the feature space to 100 dimensions. Then the t-SNE algorithm is used to reduce the 100-dimensional data to 3 dimensions, retaining the key structure of the original data. Next, a CNN model with 5 convolutional layers is used to extract deep features from the 3-dimensional data, obtaining 128 deep feature maps. Through L1 regularization feature selection, the 20 most important features are retained. The information gain score is calculated for these 20 features as the feature weight. The weight is multiplied by the feature value and summed to obtain a 20-dimensional fusion feature vector.

[0055] The fused feature vector is analyzed using a fuzzy inference system with triangular membership functions to obtain the membership values ​​of five linguistic variables ("excellent", "good", "average", "poor", and "extremely poor"). The confidence of each evaluation result is calculated, and a confidence threshold of 0.7 is set for screening. Finally, the filtered results are weighted averaged to obtain a comprehensive score from 0 to 100, where 80 points or more indicate that the pipeline is in good condition, 60-80 points indicate that close monitoring is required, and below 60 points indicate that immediate maintenance is required. In this example, the pipeline status assessment results show that the comprehensive score of the 100-meter pipe section is 72 points, which is in the range of needing close monitoring. Among them, there is a 15-meter section with a score lower than 60 points. The main problems are slight corrosion and loose joints. It is recommended to give priority to this section of the pipe in the next maintenance cycle.

[0056] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] (1) Perform parameter normalization processing on the pipeline status assessment data to obtain normalized assessment data, and perform weight allocation on the normalized assessment data to obtain weighted assessment data;

[0058] (2) Construct an indicator system for the weighted evaluation data to obtain a health indicator system, and quantify the indicators of the health indicator system to obtain quantitative indicator data;

[0059] (3) Perform fuzzy comprehensive evaluation on the quantitative indicator data to obtain fuzzy evaluation results, and then perform clarification processing on the fuzzy evaluation results to obtain a clear health index;

[0060] (4) Use the grey prediction algorithm to perform time series analysis on the clear health index to obtain health trend data, and set thresholds for the health trend data to obtain risk threshold data;

[0061] (5) Compare and analyze the risk threshold data and health trend data to obtain the risk identification results, and classify the risk identification results to obtain the initial risk level;

[0062] (6) Perform knowledge correction on the initial risk level to obtain the corrected risk level, and calculate the confidence interval of the corrected risk level to obtain the risk level.

[0063] Specifically, the normalization process usually adopts the minimum-maximum normalization method, and its formula is:

[0064] ;

[0065] in, is the original data value, and are the minimum and maximum values ​​of the parameter, respectively. is the normalized value. After this processing, all parameters are mapped to the interval [0, 1].

[0066] When assigning weights to normalized evaluation data, the commonly used method is the entropy weight method. The calculation formula of the entropy weight method is as follows:

[0067] ;

[0068] in, For the The weight of the indicator, For the The information entropy of the indicator, is the information entropy of the kth indicator, and n is the total number of indicators. Information entropy The calculation formula is:

[0069] ;

[0070] Where m is the number of samples, For the Sample No. The weight of each indicator is determined by the weighted evaluation data. The indicator system is constructed for the weighted evaluation data to obtain the health indicator system. This step usually adopts the analytic hierarchy process (AHP). The AHP method determines the relative importance of each indicator by constructing a judgment matrix and calculating the eigenvector. Indicator quantification is the process of converting qualitative indicators into quantitative indicators. Commonly used methods include expert scoring method and fuzzy evaluation method.

[0071] The quantitative index data is subjected to fuzzy comprehensive evaluation. This step uses fuzzy mathematics theory. First, the evaluation factor set is established. and rating set , and then construct the fuzzy relationship matrix R. The result B of the fuzzy comprehensive evaluation can be calculated by the following formula:

[0072] ;

[0073] in, is the weight vector, Represents a fuzzy composite operation.

[0074] The gray prediction algorithm is used to perform time series analysis on the clear health index, and the GM (1,1) model is commonly used. The basic steps of the GM (1,1) model include cumulative generation, establishing differential equations, solving differential equations, and restoring predicted values. The prediction formula is:

[0075] ;

[0076] in, represents the first value of the original sequence, represents the prediction step of the time series, For the The predicted value of the period, is the first value of the original sequence, and is the model parameter. Compare and analyze the risk threshold data and health trend data to obtain the risk discrimination result. Fuzzy set theory can be used in this step. Fuzzy set A is set to represent the health state and B is set to represent the risk state. The membership functions of the two are and Risk identification can be achieved by calculating the difference in membership:

[0077] ;

[0078] like , it is judged as a risk state; if , it is judged as healthy.

[0079] For example, suppose a 1,000-meter-long urban drainage pipe is evaluated for health status, and three indicators are selected: pipe wall thickness, crack length, and corrosion depth. After the original data is normalized, the normalized evaluation data is obtained: pipe wall thickness [0.8, 0.75, 0.9, 0.85, 0.7], crack length [0.1, 0.15, 0.05, 0.2, 0.25], corrosion depth [0.3, 0.35, 0.25, 0.4, 0.45]. The weights of the three indicators calculated using the entropy weight method are 0.4, 0.3, and 0.3 respectively. Multiply the weights by the normalized data to obtain the weighted evaluation data. Construct a three-layer health indicator system: overall health status, structural integrity, and material performance. Determine the weights of each layer of indicators using the AHP method.

[0080] The quantitative indicator data were subjected to fuzzy comprehensive evaluation, and the evaluation levels were set as "excellent", "good", "average", "poor", and "dangerous". The fuzzy evaluation results [0.1, 0.3, 0.4, 0.2, 0] were obtained by fuzzy relationship matrix calculation, and the clear health index was 75 (full score 100) after clarification. The GM (1,1) model was used to predict the health index in the past 5 years, and the health trend data for the next 3 years [73, 71, 69] were obtained. The risk threshold was set to 70, and the risk threshold data was obtained. Comparative analysis showed that risks began to appear in the second year. Risk discrimination was performed through fuzzy set theory, and the risk discrimination results were obtained. The risk level was divided into three levels: low, medium, and high, and the initial risk level was medium. Combined with expert knowledge, correction was performed, and the risk level after correction was still medium. Finally, the confidence interval was calculated by the Bootstrap method, and the 95% confidence interval of the risk level was [medium-0.5, medium+0.5].

[0081] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0082] (1) Perform correlation analysis on risk levels and pipeline health status indicators to obtain correlation feature data, and perform abnormal pattern recognition on the correlation feature data to obtain abnormal feature data;

[0083] (2) Performing time-frequency analysis on the abnormal feature data through the wavelet transform algorithm to obtain time-frequency feature data, and performing peak detection on the time-frequency feature data to obtain peak feature data;

[0084] (3) performing morphological processing on the peak feature data to obtain morphological feature data, and performing texture analysis on the morphological feature data to obtain texture feature data;

[0085] (4) Perform multi-dimensional feature extraction on the texture feature data through the support vector machine algorithm to obtain multi-dimensional defect feature data, and perform cluster analysis on the multi-dimensional defect feature data to obtain defect cluster data;

[0086] (5) Perform feature matching on the defect cluster data to obtain feature matching results, and perform pattern recognition on the feature matching results to obtain defect pattern data;

[0087] (6) Perform multi-category labeling on the defect pattern data to obtain labeled defect data, and classify and verify the labeled defect data to obtain defect type data, wherein the defect type data includes: crack data, corrosion data, deposition data, and structural deformation data.

[0088] Specifically, the correlation analysis uses the grey correlation analysis method to calculate the correlation between the risk level and each health status indicator. The higher the correlation, the greater the impact of the indicator on the risk level. The obtained correlation feature data contains the intrinsic connection between risk and health status. Subsequently, the correlation feature data is subjected to abnormal pattern recognition, and the local outlier factor (LOF) algorithm is used. The LOF algorithm identifies outliers by calculating the local density ratio of a data point to its neighboring points, thereby obtaining abnormal feature data. The abnormal feature data is subjected to time-frequency analysis by the wavelet transform algorithm to obtain time-frequency feature data. Wavelet transform can provide information in both the time domain and the frequency domain at the same time, and is particularly suitable for analyzing non-stationary signals. In pipeline detection, wavelet transform can effectively capture defect features at different scales. Peak detection is performed on the time-frequency feature data, and the continuous wavelet transform (CWT) method is used. CWT can detect the local maximum of the signal at different scales, thereby obtaining peak feature data. These peaks usually correspond to defects or abnormal points in the pipeline.

[0089] The peak feature data is morphologically processed to obtain morphological feature data. Morphological processing includes dilation and erosion operations, which are used to enhance or suppress specific image features. In pipeline inspection, morphological processing can help highlight the contours and structures of defects. Subsequently, the morphological feature data is texture analyzed using the gray-level co-occurrence matrix (GLCM) method. GLCM analysis can extract texture features such as energy, contrast, correlation, etc., which are essential for distinguishing different types of pipeline defects. Multidimensional feature extraction is performed on the texture feature data using the support vector machine (SVM) algorithm to obtain multidimensional defect feature data. SVM is a powerful machine learning algorithm that can construct the optimal classification hyperplane in a high-dimensional feature space. In this step, SVM is used to extract the most discriminative feature combination from the texture features. Cluster analysis is performed on the multidimensional defect feature data using the K-means clustering algorithm. K-means clustering classifies similar defect features into one category, thereby obtaining defect cluster data. This helps to preliminarily classify different types of pipeline defects.

[0090] The nearest neighbor algorithm is used to match the defect cluster data. The nearest neighbor algorithm finds the most similar samples by calculating the distance between feature vectors to obtain the feature matching results. This step matches unknown defects with known defect types. Subsequently, the feature matching results are subjected to pattern recognition using the random forest algorithm. Random forest is an ensemble learning method that can handle high-dimensional data and provide good classification performance. Through the random forest algorithm, defect pattern data is obtained, which contains a preliminary judgment of the defect type. The defect pattern data is labeled in multiple categories using a one-vs-all strategy. This strategy trains a binary classifier for each defect type (cracks, corrosion, deposition, and structural deformation) to achieve multi-category classification. After obtaining the labeled defect data, classification verification is performed using the cross-validation method. Cross-validation can evaluate the generalization ability of the classification model and ensure that the model can perform well on unseen data. The defect type data finally obtained include crack data, corrosion data, deposition data, and structural deformation data, which accurately reflect the various defects inside the pipeline.

[0091] For example, suppose a 1000-meter-long urban drainage pipe is inspected using three sensors: ultrasonic, electromagnetic, and video. First, through grey correlation analysis, the correlation between the risk level and the three indicators of pipe wall thickness, crack length, and corrosion depth is calculated to be 0.85, 0.78, and 0.92, respectively. This shows that the corrosion depth has the greatest impact on the risk level. The LOF algorithm is used for anomaly detection, and the threshold is set to 1.5. 50 abnormal points are identified, which are mainly distributed in the 400-600 meter section of the pipeline. Wavelet transform is performed on these 50 abnormal points, and Morlet wavelet is used for 5 scale decomposition. At the third scale, 15 significant peaks are detected, which may correspond to serious defect locations. Through morphological processing, the contour of the defect is enhanced by dilation and corrosion operations using a 5×5 structural element. GLCM is calculated for the processed data to extract five texture features such as energy, contrast, and entropy.

[0092] The texture features were extracted using the SVM with RBF kernel function to obtain a 10-dimensional feature space. In this feature space, the K-means clustering algorithm was applied, and K=4 was set, corresponding to 4 possible defect types. The clustering results showed that 6 of the 15 peaks belonged to the first category, 4 to the second category, 3 to the third category, and 2 to the fourth category. The 15 peaks were compared with the pre-established defect database by feature matching using the nearest neighbor algorithm. Subsequently, a random forest containing 100 decision trees was used for pattern recognition to obtain a preliminary defect type judgment. Finally, four SVM classifiers were trained using a one-to-many strategy, corresponding to four defect types: cracks, corrosion, deposition, and structural deformation. The model performance was evaluated using 5-fold cross validation, with an average accuracy of 92%. The final defect type data showed that in the 1,000-meter pipeline tested, 5 cracks (total length of about 2.5 meters), 3 corrosions (maximum depth of 30% of the pipe wall thickness), 4 depositions (maximum thickness of about 50 mm), and 1 structural deformation (deformation rate of about 5%) were found.

[0093] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0094] (1) The spatial coordinates of the defect type data are calculated by triangulation algorithm to obtain the defect coordinate data, and the defect coordinate data is mapped to the pipeline geometry to obtain the defect distribution data;

[0095] (2) Perform density analysis on the defect distribution data to obtain density distribution data, and perform hot spot analysis on the density distribution data to obtain clustered area data;

[0096] (3) The minimum enclosing rectangle algorithm is used to extract the boundary of the clustered area data to obtain the boundary data, and the area of ​​the boundary data is calculated to obtain the range data;

[0097] (4) performing depth measurement on the range data to obtain depth data, and performing volume estimation on the depth data to obtain volume data;

[0098] (5) Perform stress simulation on volume data through finite element analysis algorithm to obtain stress distribution data, and perform strength evaluation on the stress distribution data to obtain structural strength data;

[0099] (6) Calculate the safety factor of the structural strength data to obtain the safety assessment data, and quantify and grade the safety assessment data to obtain a quantitative data set of defect severity.

[0100] Specifically, in the pipeline internal detection method based on multi-sensor fusion, the spatial coordinates of the defect type data are first calculated using a triangulation algorithm. The triangulation algorithm is based on the following formula:

[0101] ;

[0102] in, is the distance from the target point to the camera, is the distance between the two cameras (baseline length), is the focal length of the camera, is the parallax of the same point in the two camera images. The defect coordinate data is obtained, and then the pipeline geometric mapping is performed on the defect coordinate data to obtain the defect distribution data. The geometric mapping uses the cylindrical coordinate system transformation:

[0103] ;

[0104] Where (X, Y, Z) is a point in the Cartesian coordinate system, and (r,θ, h) is the corresponding cylindrical coordinate system coordinate. Density analysis of defect distribution data is performed using the kernel density estimation method:

[0105] ;

[0106] here, is the density estimation function, is the sample size, is the bandwidth parameter, is the kernel function, is the sample point. After obtaining the density distribution data, hot spot analysis is performed using Getis-Ord Statistics:

[0107] ;

[0108] in, is the hotspot statistic, is the spatial weight, is the attribute value, is the average value, is the standard deviation, Represents the total number of samples (total number of spatial units) in the study area. This step obtains the clustered area data. The minimum enclosing rectangle algorithm is used to extract the boundaries of the clustered area data to obtain the boundary data. The minimum enclosing rectangle algorithm is based on the rotating caliper method, and the computational complexity is O(v), where v is the number of input point sets. The algorithm first calculates the convex hull of the point set, and then finds the minimum enclosing rectangle in O(v) time using the rotating caliper method. The area of ​​the boundary data is calculated to obtain the range data. The area calculation uses the polygon area formula:

[0109] ;

[0110] in, are the vertex coordinates of the polygon and M is the number of vertices.

[0111] Depth measurement is performed on the range data to obtain depth data. Depth measurement can use structured light or time-of-flight method. Then volume estimation is performed on the depth data to obtain volume data. Volume estimation uses voxelization method:

[0112] ;

[0113] here, is the total volume, is the volume of a single voxel, and m is the number of voxels. The stress distribution data is obtained by performing stress simulation on the volume data through the finite element analysis algorithm. Finite element analysis is based on the stress-strain relationship:

[0114] ;

[0115] is stress, is Young's modulus, is the strain. The stress distribution data is evaluated for strength to obtain the structural strength data. The strength evaluation adopts the von Mises yield criterion:

[0116] ;

[0117] is the von Mises stress, , , is the principal stress.

[0118] Finally, the safety factor is calculated for the structural strength data to obtain the safety assessment data:

[0119] ;

[0120] is the safety factor, is the yield strength of the material, is the maximum stress. The safety assessment data is quantitatively graded to obtain a quantitative data set of defect severity.

[0121] For example: suppose an inspection is conducted in a 1000m long, 1m diameter urban drainage pipe. Using the triangulation algorithm, a crack is found at 500m with spatial coordinates (0.4m, 0.3m, 500m). Mapping it to the pipe cylindrical coordinate system yields (0.5m, 36.87°, 500m). Density analysis shows that there are multiple defects between 490m and 510m, and hot spot analysis identifies a highly concentrated area. The minimum enclosing rectangle algorithm extracts an area of ​​0.2m². Depth measurement shows that the crack reaches 15mm at its deepest point. Volume estimation gives a defect volume of approximately 3000mm³. Finite element analysis shows that at the point of maximum stress, the von Mises stress reaches 120MPa. Assuming the yield strength of the pipe material is 240MPa, the safety factor is calculated to be 2. According to the preset classification criteria, this defect is quantified as "moderately severe" and needs to be repaired in the next maintenance cycle.

[0122] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0123] (1) Perform data fusion on the defect severity quantitative data set and risk level to obtain comprehensive evaluation data, and assign weights to the comprehensive evaluation data to obtain weighted evaluation data;

[0124] (2) The criterion system is constructed for the weighted evaluation data through the hierarchical analysis method to obtain a multi-criteria system, and the consistency of the multi-criteria system is tested to obtain a criterion system after the test;

[0125] (3) Perform fuzzy comprehensive evaluation on the post-inspection criterion system to obtain fuzzy evaluation results, and clarify the fuzzy evaluation results to obtain clear priority data;

[0126] (4) Prioritize the clear priority data through a sorting algorithm to obtain sorting result data, and perform threshold segmentation on the sorting result data to obtain hierarchical priority data;

[0127] (5) Perform repair technology matching on the hierarchical priority data to obtain technology matching data, and perform resource demand analysis on the technology matching data to obtain resource configuration data;

[0128] (6) Conduct a cost-benefit analysis on the resource allocation data to obtain cost-benefit data, and optimize the cost-benefit data to obtain repair recommendation data.

[0129] Specifically, data fusion uses the weighted summation method to integrate data from different sources into a unified evaluation index. Considering the relative importance of defect severity and risk level, comprehensive evaluation data is obtained. Subsequently, the comprehensive evaluation data is weighted, and the entropy weight method is used to determine the weight of each indicator. The entropy weight method is based on the information entropy theory and can objectively reflect the amount of information of the indicator, thereby obtaining weighted evaluation data. The criterion system of the weighted evaluation data is constructed by the hierarchical analysis method to obtain a multi-criteria system. The hierarchical analysis method decomposes complex problems into a hierarchical structure, including the target layer, the criterion layer, and the solution layer. In pipeline inspection, the target layer may be the pipeline maintenance priority, the criterion layer includes factors such as defect severity, risk level, and repair difficulty, and the solution layer is the specific pipeline section. Construct a judgment matrix and determine the relative importance of each element by pairwise comparison. Perform a consistency test on the multi-criteria system to ensure the rationality of the judgment. The consistency test is achieved by calculating the consistency ratio CR. If CR<0.1, the judgment matrix is ​​considered to have satisfactory consistency.

[0130] The fuzzy comprehensive evaluation is performed on the post-inspection criterion system. This step uses fuzzy mathematics theory to deal with the uncertainty in the evaluation process. First, the evaluation factor set and the comment set are established, then the fuzzy relationship matrix is ​​constructed, and finally a multi-level fuzzy comprehensive evaluation is performed. The fuzzy evaluation result obtained is a fuzzy vector, which represents the degree of membership of each evaluation level. The fuzzy evaluation result is clarified, and the center of gravity method is used to convert the fuzzy result into a clear numerical value to obtain clear priority data. The clear priority data is prioritized by the sorting algorithm to obtain the sorting result data. The sorting algorithm can select efficient algorithms such as quick sort or heap sort to arrange the pipeline sections in descending order according to the clear priority value. The sorting result data is threshold segmented, and multiple thresholds are set to divide the sorting results into different priority levels, such as high priority, medium priority and low priority, so as to obtain hierarchical priority data.

[0131] The hierarchical priority data is matched with repair technologies. This step selects appropriate repair technologies for each pipeline segment based on the defect type and severity. Repair technologies include lining repair, local patching, excavation and replacement, etc. By establishing a mapping relationship between defect characteristics and repair technologies, technology matching data is obtained. Resource demand analysis is performed on the technology matching data to evaluate the resources such as manpower, equipment, materials and time required for each repair technology to obtain resource allocation data. Finally, a cost-benefit analysis is performed on the resource allocation data to calculate the cost and expected benefit of each repair plan. Costs include direct repair costs and indirect costs (such as losses caused by traffic interruptions), and benefits include extending the life of the pipeline and reducing the risk of failure. Cost-benefit data is obtained by calculating the cost-benefit ratio. To optimize the cost-benefit data, optimization methods such as dynamic programming or genetic algorithms can be used to maximize the overall repair benefit under budget constraints, and finally obtain repair recommendation data.

[0132] For example, suppose a 5,000-meter-long drainage network in a city is inspected and divided into 50 100-meter-long sections. The defect severity (0-100 points) and risk level (1-5 levels) of each section are evaluated. Data fusion is performed using the weighted summation method, assuming that the weights are 0.6 and 0.4 respectively. For example, if the defect severity score of a section is 75 and the risk level is 4, then its comprehensive evaluation score is 75×0.6 + (4 / 5)×100×0.4= 77. A criterion system is constructed using the hierarchical analysis method, including three criteria: defect severity, risk level, and repair difficulty. The judgment matrix is ​​obtained through expert scoring, and the weight vector is calculated to be [0.5, 0.3, 0.2], with a consistency ratio CR=0.05<0.1, which passes the consistency test.

[0133] Perform fuzzy comprehensive evaluation and set the comment set to {very poor, poor, average, good, excellent}. The evaluation result of a certain pipe section is [0.1, 0.3, 0.4, 0.2, 0], and the priority score is 65 (out of 100) after being clarified by the centroid method. Perform quick sorting on 50 pipe sections to get the priority ranking. Set the thresholds of 75 and 50, divide the pipe sections into three priorities of high, medium and low, and get the hierarchical priority data. For example, there are 10 high priority pipe sections, 25 medium priority pipe sections, and 15 low priority pipe sections.

[0134] The high-priority pipe sections were matched with repair technologies, assuming that 5 of them were suitable for lining repair, 3 required local repair, and 2 required excavation and replacement. Analyzing resource requirements, lining repair costs 2,000 yuan per meter, local repair costs 5,000 yuan per location, and excavation and replacement costs 5,000 yuan per meter. A cost-benefit analysis was conducted, and the total repair cost was calculated to be 1,350,000 yuan. It is expected that the life of the pipeline network can be extended by 15 years, and the annual maintenance cost can be saved by 100,000 yuan. The repair plan was optimized through the dynamic programming algorithm, and the final repair recommendation was obtained under the budget limit of 1,500,000 yuan: 8 high-priority pipe sections and 5 medium-priority pipe sections were repaired immediately, and the rest were included in the next year's plan.

[0135] The embodiment of the present invention also provides a pipeline internal detection system based on multi-sensor fusion, such as Figure 2 As shown, the pipeline internal detection system based on multi-sensor fusion specifically includes:

[0136] The processing module 201 is used to pre-process the raw data inside the pipeline collected by multiple sensors to obtain a multi-source data set, and perform spatiotemporal registration on the multi-source data set to obtain a registered data set;

[0137] A fusion module 202 is used to extract and fuse the registered data set to obtain a target feature vector, and perform multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data;

[0138] The prediction module 203 is used to calculate the health index of the pipeline status assessment data to obtain the pipeline health status index, and to perform risk prediction on the pipeline health status index to obtain the risk level;

[0139] The classification module 204 is used to identify pipeline defect characteristics for the pipeline health status indicator based on the risk level to obtain defect characteristic data, and to perform multi-category classification processing on the defect characteristic data to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data;

[0140] The analysis module 205 is used to perform precise position analysis on the defect type data through a spatial positioning algorithm to obtain defect spatial distribution data, and to perform structural integrity assessment on the defect spatial distribution data to obtain a defect severity quantification data set;

[0141] The generation module 206 is used to perform a multi-criteria decision analysis on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and to generate an intelligent repair plan for the defect repair priority ranking data to obtain repair suggestion data.

[0142] Through the collaborative work of the above modules, the raw data collected by multiple sensors inside the pipeline are preprocessed and time-space registered, ensuring the consistency of data from different sensors in time and space, laying a solid foundation for subsequent analysis. This step effectively improves the data quality and reduces the errors caused by sensor differences. Secondly, by extracting and fusion of the registered data set, the target feature vector is obtained, and multi-level fusion analysis is performed, which greatly enhances the comprehensive assessment ability of the pipeline status. This multi-dimensional feature analysis makes the pipeline status assessment more comprehensive and accurate, and can capture subtle changes that may be ignored by a single sensor. Then, the health index calculation and risk prediction of the pipeline status assessment data not only provide quantitative indicators of the current pipeline health status, but also predict possible risks in the future, providing an important basis for preventive maintenance. This step significantly improves the foresight and initiative of pipeline management. In addition, defect feature recognition and multi-category classification processing of pipeline health status indicators based on risk levels are carried out to achieve accurate identification and classification of different types of defects, including cracks, corrosion, deposition and structural deformation. This detailed classification helps to formulate targeted repair strategies and improve maintenance efficiency. The spatial positioning algorithm is used to accurately analyze the defect type data and conduct a structural integrity assessment, which not only accurately locates the defect location, but also assesses the impact of the defect on the overall structure of the pipeline, providing key information for maintenance decisions. Finally, a multi-criteria decision analysis is performed on the defect severity quantitative data set and risk level, and an intelligent repair plan is generated, realizing intelligent management of the entire process from detection to repair.

[0143] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the embodiments, a person skilled in the art should understand that the specific implementation modes of the present invention can still be modified or replaced by equivalents, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A pipeline internal detection method based on multi-sensor fusion, characterized in that: include: Preprocessing the raw data inside the pipeline collected by multiple sensors to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set; Extracting and fusing the registered data set to obtain a target feature vector, and performing multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data; The health index of the pipeline status assessment data is calculated to obtain a pipeline health status index, and the risk prediction of the pipeline health status index is performed to obtain a risk level, including: performing parameter normalization processing on the pipeline status assessment data to obtain normalized assessment data, and performing weight allocation on the normalized assessment data to obtain weighted assessment data; constructing an indicator system for the weighted assessment data to obtain a health indicator system, and quantifying indicators of the health indicator system to obtain quantitative indicator data; performing fuzzy comprehensive evaluation on the quantitative indicator data to obtain a fuzzy evaluation result, and performing clarity processing on the fuzzy evaluation result to obtain a clear health index; performing time series analysis on the clear health index through a gray prediction algorithm to obtain health trend data, and performing threshold setting on the health trend data to obtain risk threshold data; performing comparative analysis on the risk threshold data and the health trend data to obtain a risk discrimination result, and performing grade classification on the risk discrimination result to obtain an initial risk level; performing knowledge correction on the initial risk level to obtain a corrected risk level, and performing confidence interval calculation on the corrected risk level to obtain a risk level; Based on the risk level, pipeline defect characteristics are identified for the pipeline health status indicator to obtain defect characteristic data, and the defect characteristic data is subjected to multi-category classification processing to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data; Performing precise position analysis on the defect type data through a spatial positioning algorithm to obtain defect spatial distribution data, and performing structural integrity evaluation on the defect spatial distribution data to obtain a defect severity quantification data set; A multi-criteria decision analysis is performed on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and an intelligent repair plan is generated for the defect repair priority ranking data to obtain repair suggestion data.

2. The pipeline internal detection method based on multi-sensor fusion according to claim 1 is characterized in that: The steps of preprocessing the raw data collected by multiple sensors inside the pipeline to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set include: Cleaning the raw data collected by multiple sensors to obtain cleaned data, and filtering the cleaned data to obtain filtered data; Performing data standardization processing on the filtered data to obtain standardized data, and performing data segmentation on the standardized data to obtain a segmented data set; Performing frequency domain conversion on the segmented data set by Fourier transform algorithm to obtain frequency domain data, and performing characteristic frequency band extraction on the frequency domain data to obtain characteristic frequency band data; Performing time stamp alignment on the characteristic frequency band data to obtain aligned data, and performing spatial coordinate transformation on the aligned data to obtain unified coordinate system data; Supplementing the unified coordinate system data by an interpolation algorithm to obtain a complete data set, and performing quality assessment on the complete data set to obtain a candidate multi-source data set; An outlier detection is performed on the candidate multi-source data set to obtain a detected data set, and a spatiotemporal consistency check is performed on the detected data set to obtain a target registration data set.

3. The pipeline internal detection method based on multi-sensor fusion according to claim 1 is characterized in that: The step of extracting and fusing the registered data set to obtain a target feature vector, and performing multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data includes: Performing multi-scale decomposition on the registered data set to obtain multi-scale feature data, and performing principal component analysis on the multi-scale feature data to obtain main feature data; Performing nonlinear mapping on the main feature data to obtain high-dimensional feature data, and performing dimensionality reduction processing on the high-dimensional feature data to obtain feature data after dimensionality reduction; Performing deep feature extraction on the feature data after dimension reduction by using a convolutional neural network algorithm to obtain deep feature data, and performing feature selection on the deep feature data to obtain key feature data; Performing feature weight calculation on the key feature data to obtain feature weight data, and performing weighted fusion on the feature weight data and the key feature data to obtain a fused feature vector; Performing multi-level analysis on the fused feature vector by using a fuzzy inference algorithm to obtain a multi-level analysis result, and performing confidence evaluation on the multi-level analysis result to obtain confidence data; The confidence data is threshold screened to obtain screened evaluation data, and the screened evaluation data is comprehensively scored to obtain pipeline status evaluation data.

4. The pipeline internal detection method based on multi-sensor fusion according to claim 1 is characterized in that: The step of identifying pipeline defect characteristics for the pipeline health status indicator based on the risk level to obtain defect characteristic data, and performing multi-category classification processing on the defect characteristic data to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data includes: Performing correlation analysis on the risk level and the pipeline health status indicator to obtain correlation feature data, and performing abnormal pattern recognition on the correlation feature data to obtain abnormal feature data; Performing time-frequency analysis on the abnormal characteristic data by wavelet transform algorithm to obtain time-frequency characteristic data, and performing peak detection on the time-frequency characteristic data to obtain peak characteristic data; Performing morphological processing on the peak feature data to obtain morphological feature data, and performing texture analysis on the morphological feature data to obtain texture feature data; Performing multi-dimensional feature extraction on the texture feature data by using a support vector machine algorithm to obtain multi-dimensional defect feature data, and performing cluster analysis on the multi-dimensional defect feature data to obtain defect cluster data; Performing feature matching on the defect cluster data to obtain feature matching results, and performing pattern recognition on the feature matching results to obtain defect pattern data; The defect mode data is annotated in multiple categories to obtain annotated defect data, and the annotated defect data is classified and verified to obtain defect type data, wherein the defect type data includes: crack data, corrosion data, deposition data and structural deformation data.

5. The pipeline internal detection method based on multi-sensor fusion according to claim 1 is characterized in that: The step of performing precise position analysis on the defect type data by using a spatial positioning algorithm to obtain defect spatial distribution data, and performing structural integrity assessment on the defect spatial distribution data to obtain a defect severity quantification data set includes: Performing spatial coordinate calculation on the defect type data by using a triangulation algorithm to obtain defect coordinate data, and performing pipeline geometric mapping on the defect coordinate data to obtain defect distribution data; Performing density analysis on the defect distribution data to obtain density distribution data, and performing hot spot analysis on the density distribution data to obtain clustered area data; Extracting the boundary of the clustered area data by a minimum circumscribed rectangle algorithm to obtain boundary data, and calculating the area of ​​the boundary data to obtain range data; Performing depth measurement on the range data to obtain depth data, and performing volume estimation on the depth data to obtain volume data; Performing stress simulation on the volume data by using a finite element analysis algorithm to obtain stress distribution data, and performing strength evaluation on the stress distribution data to obtain structural strength data; The safety factor is calculated for the structural strength data to obtain safety assessment data, and the safety assessment data is quantitatively graded to obtain a defect severity quantitative data set.

6. The pipeline internal detection method based on multi-sensor fusion according to claim 5 is characterized in that: The step of performing a multi-criteria decision analysis on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and performing intelligent repair scheme generation on the defect repair priority ranking data to obtain repair suggestion data includes: Performing data fusion on the defect severity quantification data set and the risk level to obtain comprehensive evaluation data, and performing weight allocation on the comprehensive evaluation data to obtain weighted evaluation data; Constructing a criterion system for the weighted evaluation data through the hierarchical analysis method to obtain a multi-criteria system, and performing consistency test on the multi-criteria system to obtain a post-test criterion system; Performing fuzzy comprehensive evaluation on the post-inspection criterion system to obtain a fuzzy evaluation result, and performing clarification processing on the fuzzy evaluation result to obtain clear priority data; Prioritizing the clear priority data by a sorting algorithm to obtain sorting result data, and performing threshold segmentation on the sorting result data to obtain hierarchical priority data; Performing repair technology matching on the hierarchical priority data to obtain technology matching data, and performing resource demand analysis on the technology matching data to obtain resource configuration data; A cost-benefit analysis is performed on the resource configuration data to obtain cost-benefit data, and a solution optimization is performed on the cost-benefit data to obtain repair suggestion data.

7. A pipeline interior detection system based on multi-sensor fusion, used to execute the pipeline interior detection method based on multi-sensor fusion as claimed in any one of claims 1 to 6, characterized in that: include: A processing module, used for preprocessing the raw data inside the pipeline collected by multiple sensors to obtain a multi-source data set, and performing spatiotemporal registration on the multi-source data set to obtain a registered data set; A fusion module is used to extract and fuse the features of the registration data set to obtain a target feature vector, and perform multi-level fusion analysis on the target feature vector to obtain pipeline status assessment data; The prediction module is used to calculate the health index of the pipeline status assessment data to obtain the pipeline health status index, and to perform risk prediction on the pipeline health status index to obtain the risk level, including: performing parameter normalization processing on the pipeline status assessment data to obtain normalized assessment data, and performing weight allocation on the normalized assessment data to obtain weighted assessment data; constructing an indicator system for the weighted assessment data to obtain a health indicator system, and quantifying the indicators of the health indicator system to obtain quantitative indicator data; performing fuzzy comprehensive evaluation on the quantitative indicator data to obtain a fuzzy evaluation result, and performing clarity processing on the fuzzy evaluation result to obtain a clear health index; performing time series analysis on the clear health index through a gray prediction algorithm to obtain health trend data, and performing threshold setting on the health trend data to obtain risk threshold data; performing comparative analysis on the risk threshold data and the health trend data to obtain a risk discrimination result, and performing grade classification on the risk discrimination result to obtain an initial risk level; performing knowledge correction on the initial risk level to obtain a corrected risk level, and performing confidence interval calculation on the corrected risk level to obtain a risk level; A classification module, used to identify pipeline defect characteristics for the pipeline health status indicator based on the risk level to obtain defect characteristic data, and to perform multi-category classification processing on the defect characteristic data to obtain defect type data including crack data, corrosion data, deposition data and structural deformation data; An analysis module is used to perform precise position analysis on the defect type data through a spatial positioning algorithm to obtain defect spatial distribution data, and to perform structural integrity assessment on the defect spatial distribution data to obtain a defect severity quantification data set; A generation module is used to perform a multi-criteria decision analysis on the defect severity quantification data set and the risk level to obtain defect repair priority ranking data, and to generate an intelligent repair plan for the defect repair priority ranking data to obtain repair suggestion data.

Images